Embedded Rule-based Management for Content-based DTNs - Advances in Network-Embedded Management and Applications - page 92

to be checked for transmission, and air time when transmitted. Optimum buer
size depends on such parameters as network latency (e.g., there is no benet in
keeping messages after a copy reaches destination) and network load (the rate
of arrival of new messages).
{ Subscription quality management parameters. These parameters control the re-
inforcement and aging behavior of the quality of subscriptions, and should be
tuned to the real probabilities of encounter in the network, which in turn depend
on the network density and movement patterns.
{ Views broadcast period. This parameter controls the rate of gossiping messages,
and must also match the network density and movement patterns. Too low a rate
will miss encounter opportunities, and too high a rate imposes an unnecessary
load on the network.
There are several other parameters that could be added to the basic protocol, like
remaining battery or available bandwidth. An example of such extension applied to
PRoPHET can be seen in [11]. Our management infrastructure uses all conguration
parameters in a consistent way, and thus provides a method to easily integrate and
take advantage of new variables.
The optimum values for those parameters can be hard to estimate in advance. For
example, the movement patterns and number of nodes change: there is a week/week-
end cycle, and an yearly cycle of vacations. Network patterns also change: a single
new subscription or sensor installed can dene a new ow of data that changes
the load imposed on the network. At the same time, the sensor nodes emit data
depending on local readings, and thus are hard to predict.
This leads to the need of a mechanism that would manipulate the conguration
parameters autonomously, adjusting them to optimize the algorithm's performance.
We implement this mechanism through the use of PDP (Policy Decision Point), a
general purpose policy engine. The idea is that a state machine is used to recognize
patterns that occur in a ow of events as described in [9]. A success in recognizing a
pattern triggers a corresponding action or set of actions. Those actions can be con-
guration change commands, or abstract notications that express the occurrence
of a situation. The later can be consumed by other recognizers in a hierarchical way.
4 System Overview
The DEMOS system (see Figure 1 ) consists of a set of services deployed on the
nodes, according to their function in the DEMOS network. The nodes fall under
some of the following categories:
{ Sensor Nodes. These are the nodes that collect environmental data. Those are
usually xed, and have attached sensing hardware. Dierent nodes can have
dierent sensing hardware attached.
{ Collectors. These nodes are usually placed in schools, and are the recipients of
data generated by the Sensor Nodes. Data is collected and relayed to a central
Management Station through the Internet.